Coupling optical detection by FlowCAM to automatic classification by the software Zoo/PhytoImage for automation of phytoplankton community identification

The marine micro-phytoplankton communities are characterized biweekly by Ifremer at reference sites along the French coasts since 1984 (French monitoring network Rephy). This exceptionally long-term series is very time-consuming and mobilizes numerous taxonomists. In order to partly automate phytoplankton identification, an original project was developed, combining optical detection of particles by the FlowCAM technology to image analysis and automatic classification performed by the Zoo/PhytoImage software. This approach, based on the machine learning principle, shows that quantitative discrimination of the particles from a natural sample is possible at a very fine taxonomic level, very close to the manual counting by microscopic method. In addition, an image and associate measurements are available for each detected particle, allowing both manual identification and numerical preservation of the samples. This study presents the classifier tool created from Arcachon bay samples, which can discriminate between 34 groups of particles with emphasis on the micro-phytoplanktonic taxa. Global identification rate with this classifier is 77.5%. The performance and limits of this innovative tool are compared with microscopic enumeration on the same samples.